"""每日刷新:下载最新国际比赛结果 → 重新入库(修正分组) → 重训模型 → 重跑模拟/预测。 世界杯期间每天跑一次,让积分榜/出线/淘汰赛/预测吃到最新真实战绩。""" from __future__ import annotations import io import sys import zipfile import urllib.request from pathlib import Path ROOT = Path(__file__).resolve().parents[2] RAW = ROOT / "data" / "raw" / "international_results-master" URL = "https://codeload.github.com/martj42/international_results/zip/refs/heads/master" def download_latest() -> int: """拉取最新数据集,覆盖本地 csv。返回新增已赛场次数(估算)。""" import pandas as pd before = 0 f = RAW / "results.csv" if f.exists(): before = int(pd.read_csv(f).dropna(subset=["home_score"]).shape[0]) print("下载最新数据集 ...", flush=True) data = urllib.request.urlopen(URL, timeout=120).read() zf = zipfile.ZipFile(io.BytesIO(data)) RAW.mkdir(parents=True, exist_ok=True) for name in ("results.csv", "shootouts.csv", "goalscorers.csv", "former_names.csv"): member = f"international_results-master/{name}" try: (RAW / name).write_bytes(zf.read(member)) except KeyError: pass after = int(pd.read_csv(f).dropna(subset=["home_score"]).shape[0]) print(f"已赛场次 {before} -> {after} (新增 {after - before})", flush=True) return after - before if __name__ == "__main__": import config # noqa: F401 (sets UTF-8) n_new = download_latest() force = "--force" in sys.argv if n_new <= 0 and not force: print("无新比赛,跳过重训。(--force 可强制重训)") sys.exit(0) print("重新入库 ...", flush=True) import ingest ingest.load() print("重训模型 ...", flush=True) from backtest import tune tune("2024-01-01") # 网格搜 xi + 用最优 xi 重训保存 print("重跑模拟/预测 ...", flush=True) import pipeline out = pipeline.run(20000) print(f"完成。model_run_id={out['run_id']} RPS={out['rps']}") # 热加载后端(若在运行),无需重启 try: urllib.request.urlopen("http://127.0.0.1:8000/api/reload", data=b"", timeout=30) print(">>> 后端已热加载新模型") except Exception: print(">>> 后端未运行或热加载失败(下次启动会自动读到新数据)")